cannam@95: /* cannam@95: * Copyright (c) 2003, 2007-11 Matteo Frigo cannam@95: * Copyright (c) 2003, 2007-11 Massachusetts Institute of Technology cannam@95: * cannam@95: * This program is free software; you can redistribute it and/or modify cannam@95: * it under the terms of the GNU General Public License as published by cannam@95: * the Free Software Foundation; either version 2 of the License, or cannam@95: * (at your option) any later version. cannam@95: * cannam@95: * This program is distributed in the hope that it will be useful, cannam@95: * but WITHOUT ANY WARRANTY; without even the implied warranty of cannam@95: * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the cannam@95: * GNU General Public License for more details. cannam@95: * cannam@95: * You should have received a copy of the GNU General Public License cannam@95: * along with this program; if not, write to the Free Software cannam@95: * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA cannam@95: * cannam@95: */ cannam@95: cannam@95: /* This file was automatically generated --- DO NOT EDIT */ cannam@95: /* Generated on Sun Nov 25 07:42:29 EST 2012 */ cannam@95: cannam@95: #include "codelet-rdft.h" cannam@95: cannam@95: #ifdef HAVE_FMA cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2cdft_c.native -fma -reorder-insns -schedule-for-pipeline -simd -compact -variables 4 -pipeline-latency 8 -trivial-stores -variables 32 -no-generate-bytw -n 10 -dif -sign 1 -name hc2cbdftv_10 -include hc2cbv.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 61 FP additions, 50 FP multiplications, cannam@95: * (or, 33 additions, 22 multiplications, 28 fused multiply/add), cannam@95: * 76 stack variables, 4 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "hc2cbv.h" cannam@95: cannam@95: static void hc2cbdftv_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DVK(KP618033988, +0.618033988749894848204586834365638117720309180); cannam@95: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 18)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 18), MAKE_VOLATILE_STRIDE(40, rs)) { cannam@95: V Ts, T4, TR, T1, TZ, TD, Ty, Tn, Ti, TT, T11, TJ, T15, Tr, TN; cannam@95: V TE, Tv, To, Tb, T8, Tw, Te, Tx, Th, Tt, T7, T9, T2, T3, Tc; cannam@95: V Td, Tf, Tg, T5, T6, Tu, Ta; cannam@95: T2 = LD(&(Rp[0]), ms, &(Rp[0])); cannam@95: T3 = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); cannam@95: Tc = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); cannam@95: Td = LD(&(Rm[0]), -ms, &(Rm[0])); cannam@95: Tf = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Tg = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: T5 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); cannam@95: T6 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); cannam@95: T8 = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Ts = VFMACONJ(T3, T2); cannam@95: T4 = VFNMSCONJ(T3, T2); cannam@95: Tw = VFMACONJ(Td, Tc); cannam@95: Te = VFNMSCONJ(Td, Tc); cannam@95: Tx = VFMACONJ(Tg, Tf); cannam@95: Th = VFMSCONJ(Tg, Tf); cannam@95: Tt = VFMACONJ(T6, T5); cannam@95: T7 = VFNMSCONJ(T6, T5); cannam@95: T9 = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: TR = LDW(&(W[TWVL * 8])); cannam@95: T1 = LDW(&(W[TWVL * 4])); cannam@95: TZ = LDW(&(W[TWVL * 12])); cannam@95: TD = VSUB(Tw, Tx); cannam@95: Ty = VADD(Tw, Tx); cannam@95: Tn = VSUB(Te, Th); cannam@95: Ti = VADD(Te, Th); cannam@95: Tu = VFMACONJ(T9, T8); cannam@95: Ta = VFMSCONJ(T9, T8); cannam@95: TT = LDW(&(W[TWVL * 6])); cannam@95: T11 = LDW(&(W[TWVL * 10])); cannam@95: TJ = LDW(&(W[TWVL * 16])); cannam@95: T15 = LDW(&(W[0])); cannam@95: Tr = LDW(&(W[TWVL * 2])); cannam@95: TN = LDW(&(W[TWVL * 14])); cannam@95: TE = VSUB(Tt, Tu); cannam@95: Tv = VADD(Tt, Tu); cannam@95: To = VSUB(T7, Ta); cannam@95: Tb = VADD(T7, Ta); cannam@95: { cannam@95: V TV, TF, Tz, TB, TL, Tp, Tj, Tl, T17, TA, TS, Tk, TC, TU, TK; cannam@95: V Tm, TO, TG, T12, TW, T16, TM, T10, Tq, TX, TY, T18, T19, TQ, TP; cannam@95: V T13, T14, TI, TH; cannam@95: TV = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), TD, TE)); cannam@95: TF = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), TE, TD)); cannam@95: Tz = VADD(Tv, Ty); cannam@95: TB = VSUB(Tv, Ty); cannam@95: TL = VMUL(LDK(KP951056516), VFMA(LDK(KP618033988), Tn, To)); cannam@95: Tp = VMUL(LDK(KP951056516), VFNMS(LDK(KP618033988), To, Tn)); cannam@95: Tj = VADD(Tb, Ti); cannam@95: Tl = VSUB(Tb, Ti); cannam@95: T17 = VADD(Ts, Tz); cannam@95: TA = VFNMS(LDK(KP250000000), Tz, Ts); cannam@95: TS = VZMULI(TR, VADD(T4, Tj)); cannam@95: Tk = VFNMS(LDK(KP250000000), Tj, T4); cannam@95: TC = VFNMS(LDK(KP559016994), TB, TA); cannam@95: TU = VFMA(LDK(KP559016994), TB, TA); cannam@95: TK = VFMA(LDK(KP559016994), Tl, Tk); cannam@95: Tm = VFNMS(LDK(KP559016994), Tl, Tk); cannam@95: TO = VZMUL(TN, VFMAI(TF, TC)); cannam@95: TG = VZMUL(Tr, VFNMSI(TF, TC)); cannam@95: T12 = VZMUL(T11, VFMAI(TV, TU)); cannam@95: TW = VZMUL(TT, VFNMSI(TV, TU)); cannam@95: T16 = VZMULI(T15, VFMAI(TL, TK)); cannam@95: TM = VZMULI(TJ, VFNMSI(TL, TK)); cannam@95: T10 = VZMULI(TZ, VFNMSI(Tp, Tm)); cannam@95: Tq = VZMULI(T1, VFMAI(Tp, Tm)); cannam@95: TX = VADD(TS, TW); cannam@95: TY = VCONJ(VSUB(TW, TS)); cannam@95: T18 = VADD(T16, T17); cannam@95: T19 = VCONJ(VSUB(T17, T16)); cannam@95: TQ = VCONJ(VSUB(TO, TM)); cannam@95: TP = VADD(TM, TO); cannam@95: T13 = VADD(T10, T12); cannam@95: T14 = VCONJ(VSUB(T12, T10)); cannam@95: TI = VCONJ(VSUB(TG, Tq)); cannam@95: TH = VADD(Tq, TG); cannam@95: ST(&(Rp[WS(rs, 2)]), TX, ms, &(Rp[0])); cannam@95: ST(&(Rm[WS(rs, 2)]), TY, -ms, &(Rm[0])); cannam@95: ST(&(Rp[0]), T18, ms, &(Rp[0])); cannam@95: ST(&(Rm[0]), T19, -ms, &(Rm[0])); cannam@95: ST(&(Rm[WS(rs, 4)]), TQ, -ms, &(Rm[0])); cannam@95: ST(&(Rp[WS(rs, 4)]), TP, ms, &(Rp[0])); cannam@95: ST(&(Rp[WS(rs, 3)]), T13, ms, &(Rp[WS(rs, 1)])); cannam@95: ST(&(Rm[WS(rs, 3)]), T14, -ms, &(Rm[WS(rs, 1)])); cannam@95: ST(&(Rm[WS(rs, 1)]), TI, -ms, &(Rm[WS(rs, 1)])); cannam@95: ST(&(Rp[WS(rs, 1)]), TH, ms, &(Rp[WS(rs, 1)])); cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(1, 1), cannam@95: VTW(1, 2), cannam@95: VTW(1, 3), cannam@95: VTW(1, 4), cannam@95: VTW(1, 5), cannam@95: VTW(1, 6), cannam@95: VTW(1, 7), cannam@95: VTW(1, 8), cannam@95: VTW(1, 9), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 10, XSIMD_STRING("hc2cbdftv_10"), twinstr, &GENUS, {33, 22, 28, 0} }; cannam@95: cannam@95: void XSIMD(codelet_hc2cbdftv_10) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cbdftv_10, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #else /* HAVE_FMA */ cannam@95: cannam@95: /* Generated by: ../../../genfft/gen_hc2cdft_c.native -simd -compact -variables 4 -pipeline-latency 8 -trivial-stores -variables 32 -no-generate-bytw -n 10 -dif -sign 1 -name hc2cbdftv_10 -include hc2cbv.h */ cannam@95: cannam@95: /* cannam@95: * This function contains 61 FP additions, 30 FP multiplications, cannam@95: * (or, 55 additions, 24 multiplications, 6 fused multiply/add), cannam@95: * 81 stack variables, 4 constants, and 20 memory accesses cannam@95: */ cannam@95: #include "hc2cbv.h" cannam@95: cannam@95: static void hc2cbdftv_10(R *Rp, R *Ip, R *Rm, R *Im, const R *W, stride rs, INT mb, INT me, INT ms) cannam@95: { cannam@95: DVK(KP250000000, +0.250000000000000000000000000000000000000000000); cannam@95: DVK(KP951056516, +0.951056516295153572116439333379382143405698634); cannam@95: DVK(KP587785252, +0.587785252292473129168705954639072768597652438); cannam@95: DVK(KP559016994, +0.559016994374947424102293417182819058860154590); cannam@95: { cannam@95: INT m; cannam@95: for (m = mb, W = W + ((mb - 1) * ((TWVL / VL) * 18)); m < me; m = m + VL, Rp = Rp + (VL * ms), Ip = Ip + (VL * ms), Rm = Rm - (VL * ms), Im = Im - (VL * ms), W = W + (TWVL * 18), MAKE_VOLATILE_STRIDE(40, rs)) { cannam@95: V T5, TE, Ts, Tt, TC, Tz, TH, TJ, To, Tq, T2, T4, T3, T9, Tx; cannam@95: V Tm, TB, Td, Ty, Ti, TA, T6, T8, T7, Tl, Tk, Tj, Tc, Tb, Ta; cannam@95: V Tf, Th, Tg, TF, TG, Te, Tn; cannam@95: T2 = LD(&(Rp[0]), ms, &(Rp[0])); cannam@95: T3 = LD(&(Rm[WS(rs, 4)]), -ms, &(Rm[0])); cannam@95: T4 = VCONJ(T3); cannam@95: T5 = VSUB(T2, T4); cannam@95: TE = VADD(T2, T4); cannam@95: T6 = LD(&(Rp[WS(rs, 2)]), ms, &(Rp[0])); cannam@95: T7 = LD(&(Rm[WS(rs, 2)]), -ms, &(Rm[0])); cannam@95: T8 = VCONJ(T7); cannam@95: T9 = VSUB(T6, T8); cannam@95: Tx = VADD(T6, T8); cannam@95: Tl = LD(&(Rp[WS(rs, 1)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Tj = LD(&(Rm[WS(rs, 3)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: Tk = VCONJ(Tj); cannam@95: Tm = VSUB(Tk, Tl); cannam@95: TB = VADD(Tk, Tl); cannam@95: Tc = LD(&(Rp[WS(rs, 3)]), ms, &(Rp[WS(rs, 1)])); cannam@95: Ta = LD(&(Rm[WS(rs, 1)]), -ms, &(Rm[WS(rs, 1)])); cannam@95: Tb = VCONJ(Ta); cannam@95: Td = VSUB(Tb, Tc); cannam@95: Ty = VADD(Tb, Tc); cannam@95: Tf = LD(&(Rp[WS(rs, 4)]), ms, &(Rp[0])); cannam@95: Tg = LD(&(Rm[0]), -ms, &(Rm[0])); cannam@95: Th = VCONJ(Tg); cannam@95: Ti = VSUB(Tf, Th); cannam@95: TA = VADD(Tf, Th); cannam@95: Ts = VSUB(T9, Td); cannam@95: Tt = VSUB(Ti, Tm); cannam@95: TC = VSUB(TA, TB); cannam@95: Tz = VSUB(Tx, Ty); cannam@95: TF = VADD(Tx, Ty); cannam@95: TG = VADD(TA, TB); cannam@95: TH = VADD(TF, TG); cannam@95: TJ = VMUL(LDK(KP559016994), VSUB(TF, TG)); cannam@95: Te = VADD(T9, Td); cannam@95: Tn = VADD(Ti, Tm); cannam@95: To = VADD(Te, Tn); cannam@95: Tq = VMUL(LDK(KP559016994), VSUB(Te, Tn)); cannam@95: { cannam@95: V T1c, TX, Tv, T1b, TR, T15, TL, T17, TT, T11, TW, Tu, TQ, Tr, TP; cannam@95: V Tp, T1, T1a, TO, T14, TD, T10, TK, TZ, TI, Tw, T16, TS, TY, TM; cannam@95: V TU, T1e, TN, T1d, T19, T13, TV, T18, T12; cannam@95: T1c = VADD(TE, TH); cannam@95: TW = LDW(&(W[TWVL * 8])); cannam@95: TX = VZMULI(TW, VADD(T5, To)); cannam@95: Tu = VBYI(VFNMS(LDK(KP951056516), Tt, VMUL(LDK(KP587785252), Ts))); cannam@95: TQ = VBYI(VFMA(LDK(KP951056516), Ts, VMUL(LDK(KP587785252), Tt))); cannam@95: Tp = VFNMS(LDK(KP250000000), To, T5); cannam@95: Tr = VSUB(Tp, Tq); cannam@95: TP = VADD(Tq, Tp); cannam@95: T1 = LDW(&(W[TWVL * 4])); cannam@95: Tv = VZMULI(T1, VSUB(Tr, Tu)); cannam@95: T1a = LDW(&(W[0])); cannam@95: T1b = VZMULI(T1a, VADD(TQ, TP)); cannam@95: TO = LDW(&(W[TWVL * 16])); cannam@95: TR = VZMULI(TO, VSUB(TP, TQ)); cannam@95: T14 = LDW(&(W[TWVL * 12])); cannam@95: T15 = VZMULI(T14, VADD(Tu, Tr)); cannam@95: TD = VBYI(VFNMS(LDK(KP951056516), TC, VMUL(LDK(KP587785252), Tz))); cannam@95: T10 = VBYI(VFMA(LDK(KP951056516), Tz, VMUL(LDK(KP587785252), TC))); cannam@95: TI = VFNMS(LDK(KP250000000), TH, TE); cannam@95: TK = VSUB(TI, TJ); cannam@95: TZ = VADD(TJ, TI); cannam@95: Tw = LDW(&(W[TWVL * 2])); cannam@95: TL = VZMUL(Tw, VADD(TD, TK)); cannam@95: T16 = LDW(&(W[TWVL * 10])); cannam@95: T17 = VZMUL(T16, VADD(T10, TZ)); cannam@95: TS = LDW(&(W[TWVL * 14])); cannam@95: TT = VZMUL(TS, VSUB(TK, TD)); cannam@95: TY = LDW(&(W[TWVL * 6])); cannam@95: T11 = VZMUL(TY, VSUB(TZ, T10)); cannam@95: TM = VADD(Tv, TL); cannam@95: ST(&(Rp[WS(rs, 1)]), TM, ms, &(Rp[WS(rs, 1)])); cannam@95: TU = VADD(TR, TT); cannam@95: ST(&(Rp[WS(rs, 4)]), TU, ms, &(Rp[0])); cannam@95: T1e = VCONJ(VSUB(T1c, T1b)); cannam@95: ST(&(Rm[0]), T1e, -ms, &(Rm[0])); cannam@95: TN = VCONJ(VSUB(TL, Tv)); cannam@95: ST(&(Rm[WS(rs, 1)]), TN, -ms, &(Rm[WS(rs, 1)])); cannam@95: T1d = VADD(T1b, T1c); cannam@95: ST(&(Rp[0]), T1d, ms, &(Rp[0])); cannam@95: T19 = VCONJ(VSUB(T17, T15)); cannam@95: ST(&(Rm[WS(rs, 3)]), T19, -ms, &(Rm[WS(rs, 1)])); cannam@95: T13 = VCONJ(VSUB(T11, TX)); cannam@95: ST(&(Rm[WS(rs, 2)]), T13, -ms, &(Rm[0])); cannam@95: TV = VCONJ(VSUB(TT, TR)); cannam@95: ST(&(Rm[WS(rs, 4)]), TV, -ms, &(Rm[0])); cannam@95: T18 = VADD(T15, T17); cannam@95: ST(&(Rp[WS(rs, 3)]), T18, ms, &(Rp[WS(rs, 1)])); cannam@95: T12 = VADD(TX, T11); cannam@95: ST(&(Rp[WS(rs, 2)]), T12, ms, &(Rp[0])); cannam@95: } cannam@95: } cannam@95: } cannam@95: VLEAVE(); cannam@95: } cannam@95: cannam@95: static const tw_instr twinstr[] = { cannam@95: VTW(1, 1), cannam@95: VTW(1, 2), cannam@95: VTW(1, 3), cannam@95: VTW(1, 4), cannam@95: VTW(1, 5), cannam@95: VTW(1, 6), cannam@95: VTW(1, 7), cannam@95: VTW(1, 8), cannam@95: VTW(1, 9), cannam@95: {TW_NEXT, VL, 0} cannam@95: }; cannam@95: cannam@95: static const hc2c_desc desc = { 10, XSIMD_STRING("hc2cbdftv_10"), twinstr, &GENUS, {55, 24, 6, 0} }; cannam@95: cannam@95: void XSIMD(codelet_hc2cbdftv_10) (planner *p) { cannam@95: X(khc2c_register) (p, hc2cbdftv_10, &desc, HC2C_VIA_DFT); cannam@95: } cannam@95: #endif /* HAVE_FMA */